Thursday, September 17, 2009

Floodless in SEATTLE: A Scalable Ethernet Architecture

The plug-and-play functionality of Ethernet simplifies many aspects of network configuration and its flat addressing simplifies the handling of topology changes and host mobility. However, Ethernet has many limitations over large network due to Flooding to locate an end host, Comprising spanning tree and Broadcasting when end hosts use ARP, DHCP etc.

The approaches that can improve scalability over conventional Ethernet are
Hybrid IP/Ethernet networks: Interconnecting small Ethernet LAN by routers that
- Ensure efficient and flexible use of network resources
- Use a routing table with size depends on the number of subnets not the hosts
- Create smaller broadcasting domain
but it eliminates many of the plug and play advantage of Ethernet
VLAN:
A large bridged network into several VLANs that can reduce the broadcast overhead imposed on hosts. Compared with IP, VLANs simplify mobility between bridges in the same VLAN. Again it introduces many problems
- Configuration overhead to trunk the VLAN at each participating bridge
- Insufficient efficiency since effective use of per-VLAN trees requires periodically moving the roots and rebalancing the trees (manually process), and interconnection between VLAN via IP gateways rather than shortest physical paths.

The author considered SEATTLE as a Scalable Ethernet Architecture for Large Enterprises that maintains the same configuration-free properties as Ethernet bridging, yet scales to large networks
- SEATTLE forward packets along the shortest path to the host based on host MAC address. It uses resolution to map a MAC address to a host’s location.
- The path between two hosts in a SEATTLE network does not change as other hosts join and leave the network.
- SEATTLE employs traffic driven caching of host-information and this reduces the amount of state required to deliver packets

SEATLE uses one hop network level DHT to provide these mapping of MAC address to location and IP at switches to avoid the control overhead of a separate directory infrastructure. It runs a link-state protocol to ensure that the switch can observe all other switches and route along the shortest path. Also it uses a hash function to map haost information at switch like (MAC address, Location) and (IP address, MAC Address). However, when a SEATTLE network is deployed over a wide area, forwarding lookups increases latency and makes the lookup more prone to failure. So SEATTLE can be configured hierarchically, by leveraging a multi-level, one hop DHT.

1 comment:

  1. I agree that there are some issues with their approach, in terms of scaling up in size and geographic reach.

    ReplyDelete